The filesystem cache in Linux is a feature, not a leak :)
Maybe we can try to tune its usage with some sysctl ?
--
Stéphane Desneux
Intel OTC - Vannes/FR
gpg:1CA35726/DFA9B0232EF80493AF2891FA24E3A2841CA35726
On 09/10/2014 11:57, Janusz Kozerski wrote:
> On 2014-10-09 11:42:56, José Bollo wrote:
>> Hi Janusz,
>>
>> On gio, 2014-10-09 at 11:20 +0200, Janusz Kozerski wrote:
>>> Hi Stephane,
>>>
>>> I have a question about tizen-commom kernel. I think there is some
>>> issue in this kernel. It looks like the kernel have a memory leak
>>> while
>> writing to the xattrs.
>>> Could you check it?
>>>
>>> I've wrote a simple bash script-test. This command will add a random
>>> text value to a "user.test" xattr for every file on a filesystem
>>> (these xattrs shoudn't break anything):
>>> ---
>>> find / \( -fstype rootfs -o -fstype ext4 \) -exec setfattr -n
>>> user.test1 -v $(cat /dev/urandom | tr -dc 'a-zA-Z0-9' | fold -w 228
>>> | head -n 1) '{}' \;
>>> 2>/dev/null
>>> ---
>>
>> I've just tried to do the same but I replaced setfattr with cat
>>> /dev/null
>>
>> And I also see on my image that the used memory raise from 442kB to 1265kB.
>>
>> I'm guessing that the leak is coming from the file caching of the kernel.
>>
>> In man 5 proc, I found that "echo 1 > /proc/sys/vm/drop_caches" reverts.
>> Can you try?
>
> I've tried - it's exactly like you said.
> Using cat gives even more leakage, and "echo 1 > /proc/sys/vm/drop_caches"
> helps.
>
> Do you know how to fix it?
>
> BR,
> Janusz
>
>>
>> Best regards
>> José
>>
>>
>>> Run in one console a 'top' command, look at the amount of free
>>> memory (keep this 'top' process alive). In other console run the
>>> script, and go back to the first console and look at the memory
>>> counter. On my machine after running this script there is about
>>> 100MB taken by the kernel (the "find" process itself takes no more
>>> then a few MB). And even if the "find...' command is done the memory isn't
>>> freed.
>>>
>>> What is intreresting: no mather how many times I will run the commad
>>> - the memory leak occurs only once.
>>> We're trying to find the cause in kernel source.
>>>
>>> (+ Tizen dev list)
>>>
>>> BR,
>>> Janusz Kozerski
>>>
>>> _______________________________________________
>>> Dev mailing list
>>> [email protected]
>>> https://lists.tizen.org/listinfo/dev
>
>
_______________________________________________
Dev mailing list
[email protected]
https://lists.tizen.org/listinfo/dev
